Presentation is loading. Please wait.

Presentation is loading. Please wait.

Trajectory Data Mining Dr. Yu Zheng Lead Researcher, Microsoft Research Chair Professor at Shanghai Jiao Tong University Editor-in-Chief of ACM Trans.

Similar presentations


Presentation on theme: "Trajectory Data Mining Dr. Yu Zheng Lead Researcher, Microsoft Research Chair Professor at Shanghai Jiao Tong University Editor-in-Chief of ACM Trans."— Presentation transcript:

1 Trajectory Data Mining Dr. Yu Zheng Lead Researcher, Microsoft Research Chair Professor at Shanghai Jiao Tong University Editor-in-Chief of ACM Trans. Intelligent Systems and Technology http://research.microsoft.com/en-us/people/yuzheng/

2 Paradigm of Trajectory Data Mining Yu Zheng. Trajectory Data Mining: An Overview. ACM Transactions on Intelligent Systems and Technology. 2015, vol. 6, issue 3.Trajectory Data Mining: An Overview

3 Trajectory Pattern Mining Moving Together Patterns Trajectory Clustering Sequential Patterns Periodic Patterns

4 Moving Together Patterns Discover a group of objects moving together for a certain time period Patterns – Flock – Convoy – Swarm – Traveling companion – Gathering Differences – The shape or density of a group – The number of objects in a group – The duration of a pattern

5 Moving Together Patterns Flock – a group of objects that travel together within a disc of some user-specified size for at least k consecutive timestamps – The pre-defined circular shape may not well describe the shape of a group in reality Convoy – captures generic trajectory pattern of any shape – by employing the density-based clustering – requires a group of objects to be density-connected during k consecutive time points Swarm – a cluster of objects lasting for at least k (possibly non-consecutive) timestamps Convoy and swarm need to load entire trajectories into memory for a pattern mining!

6 Moving Together Patterns k = 2

7 Trajectory Pattern Mining Moving Together Patterns Trajectory Clustering Sequential Patterns Periodic Patterns

8 Trajectory Clustering Group similar trajectories into clusters – To find representative paths or – common trends shared by different moving objects In free spaces – Distance between two entire trajectories – Distance between segments of trajectories In a road network setting – Can be converted to another problem – Map-matching + graph clustering

9 Trajectory Clustering Lee et al. [1] – partition trajectories into line segments by MDL-based method – build groups of close segments using the Trajectory- Hausdorff Distance [1] J. G. Lee, J. Han, and K. Y. Whang. Trajectory clustering: A partition-and-group framework. SIGMOD 2007

10 Trajectory Clustering Micro-and-Macro-clustering framework – First find mirco-clusters of trajectory segments – Then group micro-clusters into macro-clusters Li et al. [2] – new data will only affect the local area where the new data were received rather than the far-away areas – Incremental clustering algorithm [2] Z. Li, J. Lee, X.Li, and J. Han. Incremental Clustering for Trajectories. DASFAA 2010

11 Trajectory Pattern Mining Moving Together Patterns Trajectory Clustering Sequential Patterns Periodic Patterns

12 Mining Sequential Patterns from Trajectories A certain number of moving objects traveling a common sequence of locations in a similar time interval Applications – travel recommendation, – life pattern understanding, – next location prediction, – estimating user similarity – trajectory compression How to define a location – Exact match: Check-in, road segment ID – Approximate match: spatial closeness, GPS trajectories Free spaces or in a road network setting

13 Mining Sequential Patterns from Trajectories In a free space – Line simplification-based method Using line simplification algorithm to compress a trajectory Group simplified segments based on distance (without considering temporal gaps) – Clustering-based methods [1] Detect stay points from trajectories Clustering stay points into regions Apply PrefixSpan or CloseSpan to find sequential patterns [1] Y. Ye, Y. Zheng, et al. Mining Individual Life Pattern Based on Location History. MDM 2009

14 Mining Sequential Patterns from Trajectories

15 Trajectory Pattern Mining Moving Together Patterns Trajectory Clustering Sequential Patterns Periodic Patterns

16 Periodicity is a very common phenomenon Moving objects usually have periodic behaviors: – people: go to work and go back home every weekday – animals: migrate yearly Mining periodic behaviors is useful to: – summarize over long historical movement – predict future movement – detect abnormal event my periodic behavior: 10:00am office 1:00pm home 2:00pm office 6:00pm home 7:30pm office 11:00pm home gym, tues. & thurs. grocery, weekend Grocery Home Office Gym bald eagle: yearly migration Zhenhui Li, Bolin Ding, Jiawei Han, and Roland Kays, Mining Periodic Behaviors for Moving Objects, KDD 2010Mining Periodic Behaviors for Moving Objects

17 Find the right spot to observer the movement The concrete trajectory is not important. We can observe its movement from the hive (in or out). in hive outside hive The movement is transformed into a binary sequence (in hive or outside hive). The period in the binary sequence is easy to be detected. time Zhenhui Li, Bolin Ding, Jiawei Han, and Roland Kays, Mining Periodic Behaviors for Moving Objects, KDD 2010Mining Periodic Behaviors for Moving Objects

18 Periodica outline Step 1: Detect periods – find reference spots – for each reference spot: movement is transformed into a binary sequence detect periods in the binary sequence Step 2: Summarize periodic behaviors – for each period, segment the movement by period – hierarchically cluster segments – a behavior is summarized over the segments in a cluster Zhenhui Li, Bolin Ding, Jiawei Han, and Roland Kays, Mining Periodic Behaviors for Moving Objects, KDD 2010Mining Periodic Behaviors for Moving Objects

19 Periodica: Detect periods: find reference spots first 50 days: daily periodic behavior between nest and foraging area second 50 days: daily periodic behavior between another nest and the same foraging area Reference spot: (1) frequently visited regions/locations; (2) higher density than a random location Use kernel-based method to calculate the densities Reference spots: contours of high density places foraging area nest 1 nest 2 Running Example

20 Periodica: Detect periods: transform into in-and-out binary sequence inside ref. spot 1 outside ref. spot 1 Zhenhui Li, Bolin Ding, Jiawei Han, and Roland Kays, Mining Periodic Behaviors for Moving Objects, KDD 2010Mining Periodic Behaviors for Moving Objects

21 Periodica: Detect periods: detect periods in binary sequence inside ref. spot 1 outside ref. spot 1 Fourier transform (periodogram) will give a range of periods. [23,26] Period detected here is 24 (hours). Autocorrelation further confirms the exact periods.

22 Periodica: Summarize behaviors: segment movements using the period 2 2 2 2 2 0 2 2 2 1 1 1 1 1 0 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1 2 1 1 1 1 1 1 0 1 1 1 1... 3 3 0 3 3 3 3 3 3 1 1 1 1 1 0 1 1 1 1 1 1 1 3 3 0 3 3 3 3 3 1 1 1 0 1 1 1 1 1 0 0 1 1 1.... day 1 day 2 day n day n+1 First, the movement is symbolized using ref. spots. (0 means it is outside any ref. spot.) Given the period T=24(hours), the movement is segmented into “ day ” s. Zhenhui Li, Bolin Ding, Jiawei Han, and Roland Kays, Mining Periodic Behaviors for Moving Objects, KDD 2010Mining Periodic Behaviors for Moving Objects

23 Periodica: Summarize behaviors: hierarchically cluster segments Bottom-up hierarchical clustering. Initially, each segment is a behavior. The distance between behaviors are calculated using KL-divergence. cluster (a set of segments) = behavior = probability matrix 2 2 2 2 2 0 2 2 2 1 1 1 1 1 0 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1 2 1 1 1 1 1 1 0 1 1 1 1 2 2 2 2 0 2 2 2 2 1 1 1 0 1 1 1 1 1 1 1 1 1 2 0 2 2 2 2 2 2 2 1 1 1 1 1 1 0 1 1 1 1 1 1... 3 3 0 3 3 3 3 3 3 1 1 1 1 1 0 1 1 1 1 1 1 1 3 3 3 0 3 3 3 3 1 1 1 0 1 1 1 1 1 0 0 1 1 1 3 3 3 0 3 0 3 3 3 3 1 0 1 1 1 0 1 1 1 0 1 1 3 3 3 3 3 3 3 0 3 1 1 1 1 0 1 1 1 1 1 1 1 1.... spot1234...21222324 10000...0.751.0 2 0.751.0...0000 30000 0000 000.2500...0.25000 spot1234...21222324 10000...0.75 1.0 20000...0000 31.0 0.750.5...0000 0000.250...0.25 00 Cluster 1 Cluster 2 Zhenhui Li, Bolin Ding, Jiawei Han, and Roland Kays, Mining Periodic Behaviors for Moving Objects, KDD 2010Mining Periodic Behaviors for Moving Objects

24 Periodica: Summarize behaviors: the number of periodic behaviors Finally, two periodic behaviors are detected. Zhenhui Li, Bolin Ding, Jiawei Han, and Roland Kays, Mining Periodic Behaviors for Moving Objects, KDD 2010Mining Periodic Behaviors for Moving Objects

25 Thanks! Yu Zheng yuzheng@microsoft.com Homepage Yu Zheng. Trajectory Data Mining: An Overview.Trajectory Data Mining: An Overview ACM Transactions on Intelligent Systems and Technology. 2015, vol. 6, issue 3.


Download ppt "Trajectory Data Mining Dr. Yu Zheng Lead Researcher, Microsoft Research Chair Professor at Shanghai Jiao Tong University Editor-in-Chief of ACM Trans."

Similar presentations


Ads by Google